Don’t compute the answer — BREED it. Start with a population of random guesses, keep the fittest, cross them to make offspring, mutate a little, repeat. Generation by generation the population climbs the fitness landscape toward the peak, exactly as evolution does, with no gradient required. Slide the generations and watch the swarm converge on the best.
A genetic algorithm evolves a population of candidate solutions through SELECTION (keep the fittest), CROSSOVER (recombine parents), and MUTATION (small random change), iterating over generations. It is a gradient-free, population-based search that trades exploitation (selection) against exploration (mutation), suited to rugged, discrete, or black-box landscapes. Over generations the best and mean fitness rise toward an optimum. A fail-loud self-check throws unless the best fitness improves across generations. ◆ real evolutionary computation, node-verified.
Convergence is stochastic and can stall on deceptive landscapes; selection pressure and mutation rate must be tuned (too greedy → premature convergence). The fitness-improves-over-generations behaviour and the operators are exact.